The reliability of the method was demonstrated through a series of
examples, each increasing the complexity in terms of the observed
scene and 
, the sample spacing. Scene I demonstrated the
feasibility of the method, and considered performance under a variety
of parameterisations. Pose estimation using only the edge
distribution was also considered, but demonstrated some difficulty at
estimating pose from the ``appearance'' of the edges. Applying the
method to Scene II demonstrated the effects of reducing the sampling
density and provided a slightly more complex scene, with excellent
results. Pose estimation with Scene III demonstrated that the method
can be extended to a larger, more realistic environment with good
results. In addition, some key problems were identified for
implementing the method in a working environment. Scene IV attempted
to tackle some of the problems identified in Scene III, particularly
that of obtaining reliable ground truth. The results of this
experiment were very good. In addition, Scene IV was used to
demonstrate the reliability of the method under changes in the scene -
an aspect which gives the method a significant advantage over many
other localisation solutions, particularly those that train neural
networks using global image statistics. Finally, Scene IV was used to
experiment with a consistency measure which can be used to
recover an unknown orientation given a database which is trained in a
fixed direction.